Capolupo et al. (2026) Multivariate geospatial modelling of drought exposure using multi-year remote sensing data

Identification

• Journal: Procedia Computer Science
• Year: 2026
• Date: 2026-01-01
• Authors: Alessandra Capolupo, Eufemia Tarantino
• DOI: 10.1016/j.procs.2026.02.134

Research Groups

• Department of Civil, Environmental, Land, Construction and Chemistry (DICATECh), Politecnico di Bari, Bari, Italy

Short Summary

This study developed a geomatics-based approach integrating multi-source satellite and meteorological data to analyze monthly drought variability in a Southern Italian vineyard over a decade, identifying three distinct drought stress classes (wet, transitional, dry) to support climate-resilient vineyard management.

Objective

• To develop and apply a geomatics-based methodology, integrating multi-source satellite and agro-climatic data, to analyze and classify monthly spatiotemporal drought variability in a Mediterranean vineyard over a decade (2015-2025) for enhanced climate resilience and precision viticulture.

Study Configuration

• Spatial Scale: A 2.1 hectare Aglianico grapevine (Taurasi biotype) vineyard in Fontanarosa, Avellino province, Campania region, Southern Italy.
• Temporal Scale: Monthly drought variability analysis over ten years (January 2015 to May 2025).

Methodology and Data

• Models used:
  • Principal Component Analysis (PCA) for dimensionality reduction and Composite Drought Index (CDI) construction.
  • K-means clustering for classifying drought stress levels.
  • Google Earth Engine (GEE) for satellite data acquisition, preprocessing (spatial filtering, cloud masking, temporal compositing), and index calculation.
  • R programming environment for statistical and multivariate analyses.
  • Temperature Vegetation Dryness Index (TVDI) for soil moisture estimation.
  • Linear models for dry and wet edge fitting in TVDI derivation.
  • Planck function for Land Surface Temperature (LST) derivation.
• Data sources:
  • Satellite data:
    • Sentinel-2 imagery (optical data: visible, near-infrared, red-edge bands; 10 m, 20 m, 60 m spatial resolution).
    • Landsat-8 imagery (visible, near-infrared, shortwave infrared, panchromatic, thermal infrared bands; 30 m, 15 m, 100 m spatial resolution).
  • Reanalysis data: AgERA5 dataset (gridded meteorological product based on ERA5 reanalysis by ECMWF).
    • Variables: Air temperature at 2 meters (K), dewpoint temperature at 2 meters (K), relative humidity at 2 meters (%), wind speed at 10 meters (m/s), precipitation rate (mm/day), surface vapor pressure (hPa), incoming solar radiation (J/m²/day).
  • Derived indices/variables:
    • Vegetation Indices (VIs): Normalized Difference Vegetation Index (NDVI), Optimized Soil-Adjusted Vegetation Index (OSAVI), Green NDVI (GNDVI), Normalized Difference Red Edge Index (NDRE).
    • Land Surface Temperature (LST).
    • Soil moisture conditions (approximated via TVDI).

Main Results

• The Composite Drought Index (CDI) was constructed from the first three principal components, which collectively explained approximately 84.5% of the total variance.
• The CDI exhibited a bimodal probability density distribution, indicating two prevalent regimes corresponding to moist/non-stressed and mild-to-moderate drought conditions.
• Unsupervised K-means clustering identified three distinct drought stress classes:
  • Wet: Mean CDI of 1.57 (range 0.92 to 2.60), characterized by high vegetation indices, cooler LST, and elevated atmospheric moisture.
  • Dry: Mean CDI of -1.54 (range -2.26 to -0.86), associated with elevated LST, low relative humidity, and depressed vegetation indices.
  • Transitional: Mean CDI of -0.086 (range -0.75 to 0.62), showing mixed environmental variable values.
• Seasonal patterns revealed that dry events predominantly occurred during summer, wet conditions prevailed in winter, and transitional phases were common in spring and autumn.
• Significant interannual variability in the frequency of each drought stress class was observed, highlighting the influence of interannual climate variability.

Contributions

• Introduced a robust, reproducible, and scalable geomatics-based methodology for detecting and classifying drought dynamics in Mediterranean vineyards.
• Developed a fully data-driven Composite Drought Index (CDI) by integrating multi-source satellite and agro-climatic variables via PCA, weighted by explained variance, providing a synthetic and interpretable metric for drought exposure.
• Applied unsupervised clustering to CDI values to identify and characterize three distinct drought regimes (Dry, Transitional, Wet) at the vineyard scale over a decadal period, offering an early warning signal for stress conditions.
• Demonstrated the applicability of unsupervised statistical approaches for viticultural drought monitoring, aligning with precision viticulture and Industry 4.0 frameworks.
• Provided actionable insights for site-specific vineyard management decisions (e.g., irrigation, mulching, phytosanitary treatments) to enhance climate resilience.

Funding

• European Union – NextGenerationUE, Progetto PRIN-PNRR WIN-RIESCO (CUP Master: F53D23009390001 – CUP: D53D23017750001-PNRR - Missione 4 “Istruzione e Ricerca” - Componente 2 “Dalla Ricerca all’Impresa” Investimento 1.1 “Fondo per il Programma Nazionale di Ricerca e Progetti di Rilevante Interesse Nazionale (PRIN)”).

Citation

@article{Capolupo2026Multivariate,
  author = {Capolupo, Alessandra and Tarantino, Eufemia},
  title = {Multivariate geospatial modelling of drought exposure using multi-year remote sensing data},
  journal = {Procedia Computer Science},
  year = {2026},
  doi = {10.1016/j.procs.2026.02.134},
  url = {https://doi.org/10.1016/j.procs.2026.02.134}
}